Add files using upload-large-folder tool

README.md

@@ -1,3 +1,67 @@
 ---
-license: mit
+pretty_name: NEExT Egonet Experiment Datasets
+tags:
+  - graph-machine-learning
+  - node-classification
+  - egonet
+  - networks
+viewer: false
 ---
+# NEExT Egonet Experiment Datasets
+
+Single-graph network datasets with per-node class labels, curated for egonet-based
+node classification with [NEExT](https://github.com/AnomalyPoint/NEExT): each node's
+k-hop egonet becomes a subgraph, subgraphs are embedded, and nodes are classified by
+their egonet embeddings.
+
+Every dataset folder contains:
+
+- `card.md` — information card: task, stats, source, license, citation, caveats
+- `metadata.json` — machine-readable: source URLs, sha256 checksums, per-graph stats, conversion details
+- `neext/` — NEExT-ready tables per graph: `edges.csv`/`nodes.csv` and Parquet mirrors
+  (`edges.parquet`/`nodes.parquet`; for coauthor-cs, coauthor-physics and reddit-graphsage the
+  Parquet nodes include full feature matrices omitted from CSV for size)
+- `source/` — original upstream files, unmodified, hash-recorded
+
+Table contract: `edges` (`src_node_id`,`dest_node_id`) + `nodes` (`node_id`, `<label>`, features...),
+int64 node IDs, undirected deduplicated edges, fully-populated label column.
+
+**Licensing**: each dataset keeps its upstream license/terms — see the per-dataset `card.md`.
+This repository is a research mirror; cite the original authors listed in each card.
+
+## Datasets
+| dataset | task | band | graphs | nodes | edges | classes | label | license |
+|---|---|---|---|---|---|---|---|---|
+| [actor](actor/card.md) | Actor category classification | medium | 1 | 7,600 | 26,659 | 5 | `actor_class` | Not stated (geom-gcn repo) |
+| [airports](airports/card.md) | Airport activity-level classification | small | 3 | 1,720 | 20,595 | 4 | `activity_quartile` | MIT (struc2vec repository) |
+| [amazon-computers](amazon-computers/card.md) | Product category classification (co-purchase) | medium | 1 | 13,752 | 245,861 | 10 | `category` | MIT (shchur/gnn-benchmark packaging) |
+| [amazon-photo](amazon-photo/card.md) | Product category classification (co-purchase) | medium | 1 | 7,650 | 119,081 | 8 | `category` | MIT (shchur packaging) |
+| [amazon-ratings](amazon-ratings/card.md) | Product rating-class prediction | medium | 1 | 24,492 | 93,050 | 5 | `rating_class` | MIT (yandex-research) |
+| [bitcoin-otc](bitcoin-otc/card.md) | Fraudulent-user detection (derived labels) | small | 1 | 5,881 | 21,492 | 3 | `trust_label` | SNAP research use; cite |
+| [blogcatalog](blogcatalog/card.md) | Blogger interest-group classification | medium | 1 | 7,460 | 131,034 | 38 | `group` | deepwalk repo GPL-3.0; data from ASU social computing repository |
+| [books](books/card.md) | Outlier book detection (Amazon co-purchase) | small | 1 | 1,418 | 3,695 | 2 | `is_outlier` | MIT (pygod-team/data) |
+| [citeseer](citeseer/card.md) | Paper topic classification (citation network) | small | 1 | 3,312 | 4,536 | 6 | `subject` | LINQS research distribution |
+| [coauthor-cs](coauthor-cs/card.md) | Research-field classification (co-authorship) | medium | 1 | 18,333 | 81,894 | 15 | `field` | MIT (shchur packaging) |
+| [coauthor-physics](coauthor-physics/card.md) | Research-field classification (co-authorship) | medium | 1 | 34,493 | 247,962 | 5 | `field` | MIT (shchur packaging) |
+| [cora](cora/card.md) | Paper topic classification (citation network) | small | 1 | 2,708 | 5,278 | 7 | `subject` | LINQS research distribution |
+| [deezer-europe](deezer-europe/card.md) | User gender classification | medium | 1 | 28,281 | 92,752 | 2 | `gender` | SNAP; cite FEATHER |
+| [disney](disney/card.md) | Outlier movie detection (co-purchase) | small | 1 | 124 | 335 | 2 | `is_outlier` | MIT (pygod-team/data) |
+| [email-eu-core](email-eu-core/card.md) | Department classification from email traffic | small | 1 | 1,005 | 16,064 | 42 | `department` | SNAP research use; cite the source papers |
+| [enron](enron/card.md) | Email spam detection | medium | 1 | 13,533 | 176,987 | 2 | `is_outlier` | MIT (pygod-team/data) |
+| [facebook-page-page](facebook-page-page/card.md) | Page category classification | medium | 1 | 22,470 | 170,823 | 4 | `page_type` | MUSAE repo GPL-3.0; cite MUSAE |
+| [github-developers](github-developers/card.md) | Web vs ML developer classification | medium | 1 | 37,700 | 289,003 | 2 | `ml_developer` | SNAP / MUSAE; cite MUSAE |
+| [lastfm-asia](lastfm-asia/card.md) | User country classification | medium | 1 | 7,624 | 27,806 | 18 | `country` | SNAP; cite FEATHER |
+| [minesweeper](minesweeper/card.md) | Mine prediction on a synthetic grid | medium | 1 | 10,000 | 39,402 | 2 | `is_mine` | MIT (yandex-research) |
+| [ogbn-arxiv](ogbn-arxiv/card.md) | arXiv subject-area classification | large | 1 | 169,343 | 1,157,799 | 40 | `arxiv_category` | ODC-BY |
+| [polblogs](polblogs/card.md) | Political-leaning classification | small | 1 | 1,490 | 16,715 | 2 | `leaning` | Research use; cite Adamic & Glance |
+| [pubmed](pubmed/card.md) | Paper topic classification (citation network) | medium | 1 | 19,717 | 44,324 | 3 | `diabetes_type` | LINQS research distribution |
+| [questions](questions/card.md) | User churn prediction | medium | 1 | 48,921 | 153,540 | 2 | `is_active` | MIT (yandex-research) |
+| [reddit-graphsage](reddit-graphsage/card.md) | Subreddit classification of posts | large | 1 | 232,965 | 11,606,919 | 41 | `subreddit` | SNAP research use; cite GraphSAGE |
+| [reddit-pygod](reddit-pygod/card.md) | Banned-user detection | medium | 1 | 10,984 | 78,516 | 2 | `is_outlier` | MIT (pygod-team/data) |
+| [roman-empire](roman-empire/card.md) | Syntactic-role classification | medium | 1 | 22,662 | 32,927 | 18 | `syntactic_role` | MIT (yandex-research) |
+| [tolokers](tolokers/card.md) | Banned crowdworker prediction | medium | 1 | 11,758 | 519,000 | 2 | `is_banned` | MIT (yandex-research) |
+| [twitch](twitch/card.md) | Explicit-content streamer classification | small-medium | 6 | 34,118 | 429,113 | 2 | `mature` | SNAP / MUSAE; GPL-3.0 code, cite MUSAE |
+| [webkb](webkb/card.md) | University web-page classification | small | 3 | 617 | 1,006 | 5 | `page_class` | Not stated (geom-gcn repo) |
+| [weibo](weibo/card.md) | Social spam detection | medium | 1 | 8,405 | 377,271 | 2 | `is_outlier` | MIT (pygod-team/data) |
+| [wikics](wikics/card.md) | CS article branch classification | medium | 1 | 11,701 | 215,603 | 10 | `category` | MIT (dataset code); article text CC BY-SA |
+| [wikipedia-articles](wikipedia-articles/card.md) | Traffic-level classification (binned) | small-medium | 3 | 19,109 | 400,497 | 5 | `traffic_quintile` | SNAP / MUSAE; cite MUSAE |

amazon-computers/card.md

@@ -0,0 +1,31 @@
+# Amazon Computers
+
+**Task**: Product category classification (co-purchase)  
+**Size band**: medium · **Label type**: semantic  
+**Label column**: `category` · **Converter**: `converters/convert_npz.py`
+
+Amazon co-purchase graph of computer products; 10 category classes, 767 binary bag-of-words review features.
+
+## Converted graphs (neext/)
+
+| graph | nodes | edges | classes | feature cols | isolated | class counts |
+|---|---|---|---|---|---|---|
+| default | 13,752 | 245,861 | 10 | 767 | 281 | 4: 5,158, 8: 2,156, 1: 2,142, 2: 1,414, 7: 818, 3: 542, … |
+
+*Conversion notes*: shchur gnn-benchmark npz; CSR adjacency symmetrized. 767 binary w_* feature columns.
+
+## Source
+
+- [amazon_electronics_computers.npz](https://raw.githubusercontent.com/shchur/gnn-benchmark/master/data/npz/amazon_electronics_computers.npz) — 31,921,488 bytes, sha256 `736ba1d9fd85eac2…`, fetched 2026-07-23
+
+**License**: MIT (shchur/gnn-benchmark packaging)  
+**Citation**: Shchur, Mumme, Bojchevski, Günnemann. Pitfalls of Graph Neural Network Evaluation. R2L @ NeurIPS 2018.
+
+- https://github.com/shchur/gnn-benchmark
+
+## Caveats
+
+- Contains isolated nodes; largest-component filtering will drop some labeled nodes.
+
+---
+*Generated by converters/make_cards.py; stats from metadata.json. Raw files: `source/`. NEExT tables: `neext/`.*

amazon-computers/metadata.json

@@ -0,0 +1,50 @@
+{
+  "name": "amazon-computers",
+  "sources": [
+    {
+      "url": "https://raw.githubusercontent.com/shchur/gnn-benchmark/master/data/npz/amazon_electronics_computers.npz",
+      "final_url": "https://raw.githubusercontent.com/shchur/gnn-benchmark/master/data/npz/amazon_electronics_computers.npz",
+      "filename": "amazon_electronics_computers.npz",
+      "bytes": 31921488,
+      "sha256": "736ba1d9fd85eac2da83a5ad25c463f04b0031a6ca01563f90a7099d9e8ffb2c",
+      "downloaded_at": "2026-07-23T02:14:48+00:00"
+    }
+  ],
+  "conversion": {
+    "script": "convert_npz.py",
+    "label_column": "category",
+    "notes": "shchur gnn-benchmark npz; CSR adjacency symmetrized. 767 binary w_* feature columns.",
+    "parquet": true
+  },
+  "graphs": {
+    "default": {
+      "nodes": 13752,
+      "edges": 245861,
+      "label_column": "category",
+      "classes": 10,
+      "class_counts": {
+        "4": 5158,
+        "8": 2156,
+        "1": 2142,
+        "2": 1414,
+        "7": 818,
+        "3": 542,
+        "6": 487,
+        "0": 436,
+        "5": 308,
+        "9": 291
+      },
+      "isolated_nodes": 281,
+      "feature_columns": 767
+    }
+  },
+  "title": "Amazon Computers",
+  "band": "medium",
+  "label_type": "semantic",
+  "task": "Product category classification (co-purchase)",
+  "license": "MIT (shchur/gnn-benchmark packaging)",
+  "citation": "Shchur, Mumme, Bojchevski, G\u00fcnnemann. Pitfalls of Graph Neural Network Evaluation. R2L @ NeurIPS 2018.",
+  "links": [
+    "https://github.com/shchur/gnn-benchmark"
+  ]
+}

amazon-photo/card.md

@@ -0,0 +1,31 @@
+# Amazon Photo
+
+**Task**: Product category classification (co-purchase)  
+**Size band**: medium · **Label type**: semantic  
+**Label column**: `category` · **Converter**: `converters/convert_npz.py`
+
+Amazon co-purchase graph of photo products; 8 classes, 745 binary features. Sibling of Amazon Computers.
+
+## Converted graphs (neext/)
+
+| graph | nodes | edges | classes | feature cols | isolated | class counts |
+|---|---|---|---|---|---|---|
+| default | 7,650 | 119,081 | 8 | 745 | 115 | 6: 1,941, 1: 1,686, 3: 915, 4: 882, 5: 823, 2: 703, … |
+
+*Conversion notes*: shchur gnn-benchmark npz; CSR adjacency symmetrized. 745 binary w_* feature columns.
+
+## Source
+
+- [amazon_electronics_photo.npz](https://raw.githubusercontent.com/shchur/gnn-benchmark/master/data/npz/amazon_electronics_photo.npz) — 17,113,960 bytes, sha256 `bdb1feb8e6ff42ee…`, fetched 2026-07-23
+
+**License**: MIT (shchur packaging)  
+**Citation**: Shchur et al. Pitfalls of GNN Evaluation. 2018.
+
+- https://github.com/shchur/gnn-benchmark
+
+## Caveats
+
+- Contains isolated nodes.
+
+---
+*Generated by converters/make_cards.py; stats from metadata.json. Raw files: `source/`. NEExT tables: `neext/`.*

amazon-photo/metadata.json

@@ -0,0 +1,48 @@
+{
+  "name": "amazon-photo",
+  "sources": [
+    {
+      "url": "https://raw.githubusercontent.com/shchur/gnn-benchmark/master/data/npz/amazon_electronics_photo.npz",
+      "final_url": "https://raw.githubusercontent.com/shchur/gnn-benchmark/master/data/npz/amazon_electronics_photo.npz",
+      "filename": "amazon_electronics_photo.npz",
+      "bytes": 17113960,
+      "sha256": "bdb1feb8e6ff42ee44024b04479145029c563fd42fc31bff80af20887ba0439a",
+      "downloaded_at": "2026-07-23T02:14:49+00:00"
+    }
+  ],
+  "conversion": {
+    "script": "convert_npz.py",
+    "label_column": "category",
+    "notes": "shchur gnn-benchmark npz; CSR adjacency symmetrized. 745 binary w_* feature columns.",
+    "parquet": true
+  },
+  "graphs": {
+    "default": {
+      "nodes": 7650,
+      "edges": 119081,
+      "label_column": "category",
+      "classes": 8,
+      "class_counts": {
+        "6": 1941,
+        "1": 1686,
+        "3": 915,
+        "4": 882,
+        "5": 823,
+        "2": 703,
+        "0": 369,
+        "7": 331
+      },
+      "isolated_nodes": 115,
+      "feature_columns": 745
+    }
+  },
+  "title": "Amazon Photo",
+  "band": "medium",
+  "label_type": "semantic",
+  "task": "Product category classification (co-purchase)",
+  "license": "MIT (shchur packaging)",
+  "citation": "Shchur et al. Pitfalls of GNN Evaluation. 2018.",
+  "links": [
+    "https://github.com/shchur/gnn-benchmark"
+  ]
+}

amazon-ratings/card.md

@@ -0,0 +1,27 @@
+# Amazon Ratings
+
+**Task**: Product rating-class prediction  
+**Size band**: medium · **Label type**: semantic  
+**Label column**: `rating_class` · **Converter**: `converters/convert_npz.py`
+
+Amazon product co-purchase graph where the label is the product's average rating bucket (5 classes); 300-dim text features. Heterophilous benchmark.
+
+## Converted graphs (neext/)
+
+| graph | nodes | edges | classes | feature cols | isolated | class counts |
+|---|---|---|---|---|---|---|
+| default | 24,492 | 93,050 | 5 | 300 | 0 | 1: 9,010, 0: 6,560, 2: 5,678, 3: 2,183, 4: 1,061 |
+
+*Conversion notes*: yandex heterophilous-graphs npz; edges symmetrized; 300 f_* feature columns.
+
+## Source
+
+- [amazon_ratings.npz](https://raw.githubusercontent.com/yandex-research/heterophilous-graphs/main/data/amazon_ratings.npz) — 27,744,018 bytes, sha256 `4c3a3e3b9d9f6cba…`, fetched 2026-07-23
+
+**License**: MIT (yandex-research)  
+**Citation**: Platonov et al. ICLR 2023.
+
+- https://github.com/yandex-research/heterophilous-graphs
+
+---
+*Generated by converters/make_cards.py; stats from metadata.json. Raw files: `source/`. NEExT tables: `neext/`.*

amazon-ratings/metadata.json

@@ -0,0 +1,45 @@
+{
+  "name": "amazon-ratings",
+  "sources": [
+    {
+      "url": "https://raw.githubusercontent.com/yandex-research/heterophilous-graphs/main/data/amazon_ratings.npz",
+      "final_url": "https://raw.githubusercontent.com/yandex-research/heterophilous-graphs/main/data/amazon_ratings.npz",
+      "filename": "amazon_ratings.npz",
+      "bytes": 27744018,
+      "sha256": "4c3a3e3b9d9f6cba0fede4625a00aad8c5721c1a36ed771367f446763241c7dd",
+      "downloaded_at": "2026-07-23T02:14:53+00:00"
+    }
+  ],
+  "conversion": {
+    "script": "convert_npz.py",
+    "label_column": "rating_class",
+    "notes": "yandex heterophilous-graphs npz; edges symmetrized; 300 f_* feature columns.",
+    "parquet": true
+  },
+  "graphs": {
+    "default": {
+      "nodes": 24492,
+      "edges": 93050,
+      "label_column": "rating_class",
+      "classes": 5,
+      "class_counts": {
+        "1": 9010,
+        "0": 6560,
+        "2": 5678,
+        "3": 2183,
+        "4": 1061
+      },
+      "isolated_nodes": 0,
+      "feature_columns": 300
+    }
+  },
+  "title": "Amazon Ratings",
+  "band": "medium",
+  "label_type": "semantic",
+  "task": "Product rating-class prediction",
+  "license": "MIT (yandex-research)",
+  "citation": "Platonov et al. ICLR 2023.",
+  "links": [
+    "https://github.com/yandex-research/heterophilous-graphs"
+  ]
+}

bitcoin-otc/card.md

@@ -0,0 +1,45 @@
+# Bitcoin OTC Trust Network
+
+**Task**: Fraudulent-user detection (derived labels)  
+**Size band**: small · **Label type**: anomaly  
+**Label column**: `trust_label` · **Converter**: `converters/convert_bitcoin_otc.py`
+
+Who-trusts-whom network of Bitcoin OTC traders with signed edge ratings (-10..+10). The dataset has no native node classes; the 3-class `trust_label` (benign / fraudulent / unknown) is DERIVED using the standard Recursive Web-of-Trust recipe from the signed-trust literature — evaluate classifiers on the labeled subset only, while egonets use the full graph.
+
+## Labeling methodology (decision approved 2026-07-23)
+
+Follows the REV2 (Kumar et al., WSDM 2018) / Fraudar ground-truth recipe as operationalized for the anonymized SNAP release (cf. Xue et al., TAS-GNN):
+
+1. Ratings are normalized from -10..+10 to [-1, +1] (divide by 10).
+2. **Seeds** — top-5 PageRank nodes of the positive-edge subgraph, a proxy for the platform founders, which the anonymized release hides. (Seed node 35 is the most prolific rater, consistent with the OTC founder account.)
+3. **Benign** — recursive propagation: a user is Benign if ANY already-Benign user rated them >= +0.5 normalized (raw >= +5). Repeat to fixpoint (the 'trusted community').
+4. **Fraudulent** — a non-benign user rated <= -0.5 normalized (raw <= -5) by ANY verified Benign user. Negative ratings from non-benign raters are ignored to prevent retaliation/bad-mouthing false positives.
+5. **Unknown** — everyone else; kept in the graph for structure, masked during supervised evaluation.
+
+Result: 636 benign / 614 fraudulent / 4,631 unknown. Parameters (top-5 PageRank seeds, +/-0.5 thresholds) and seed IDs are recorded in metadata.json under `conversion.labeling_rule`; the rule is implemented in `converters/convert_bitcoin_otc.py`.
+
+## Converted graphs (neext/)
+
+| graph | nodes | edges | classes | feature cols | isolated | class counts |
+|---|---|---|---|---|---|---|
+| default | 5,881 | 21,492 | 3 | 0 | 0 | unknown: 4,631, benign: 636, fraudulent: 614 |
+
+*Conversion notes*: Derived 3-class trust_label (benign/fraudulent/unknown) via the recursive Web-of-Trust recipe; evaluate classifiers on the labeled subset only, egonets use the full graph. Rated-pair structure symmetrized; rating signs/weights and timestamps not carried into edges. Label-only nodes.csv: rating aggregates would leak the derived label.
+
+## Source
+
+- [soc-sign-bitcoinotc.csv.gz](https://snap.stanford.edu/data/soc-sign-bitcoinotc.csv.gz) — 397,258 bytes, sha256 `6424ac981dad3a01…`, fetched 2026-07-23
+
+**License**: SNAP research use; cite  
+**Citation**: Kumar, Spezzano, Subrahmanian, Faloutsos. Edge Weight Prediction in Weighted Signed Networks. ICDM 2016. Labels: Kumar et al. REV2, WSDM 2018.
+
+- https://snap.stanford.edu/data/soc-sign-bitcoin-otc.html
+
+## Caveats
+
+- Labels are DERIVED, not oracle ground truth — comparable to literature practice, not to law-enforcement labels.
+- Evaluate on the labeled subset only (mask 'unknown'); do not treat 'unknown' as a third real class.
+- Edge sign/weight and timestamps are not carried into edges.csv (NEExT graphs are unweighted); the signs were consumed by the labeling rule.
+
+---
+*Generated by converters/make_cards.py; stats from metadata.json. Raw files: `source/`. NEExT tables: `neext/`.*

bitcoin-otc/metadata.json

@@ -0,0 +1,60 @@
+{
+  "name": "bitcoin-otc",
+  "sources": [
+    {
+      "url": "https://snap.stanford.edu/data/soc-sign-bitcoinotc.csv.gz",
+      "final_url": "https://snap.stanford.edu/data/soc-sign-bitcoinotc.csv.gz",
+      "filename": "soc-sign-bitcoinotc.csv.gz",
+      "bytes": 397258,
+      "sha256": "6424ac981dad3a019f697fc1b9fcd85c19d8d9f039797758e9ffb6fea100c373",
+      "downloaded_at": "2026-07-23T02:14:57+00:00"
+    }
+  ],
+  "title": "Bitcoin OTC Trust Network",
+  "band": "small",
+  "label_type": "anomaly",
+  "task": "Fraudulent-user detection (derived labels)",
+  "license": "SNAP research use; cite",
+  "citation": "Kumar, Spezzano, Subrahmanian, Faloutsos. Edge Weight Prediction in Weighted Signed Networks. ICDM 2016. Labels: Kumar et al. REV2, WSDM 2018.",
+  "links": [
+    "https://snap.stanford.edu/data/soc-sign-bitcoin-otc.html"
+  ],
+  "conversion": {
+    "script": "convert_bitcoin_otc.py",
+    "label_column": "trust_label",
+    "labeling_rule": {
+      "method": "Recursive Web-of-Trust (REV2/Fraudar recipe, TAS-GNN operationalization)",
+      "normalization": "rating / 10 -> [-1, +1]",
+      "seeds": "top-5 PageRank nodes of the positive subgraph (founder proxies)",
+      "benign": "recursive: any rating >= +0.5 (raw +5) from a Benign node",
+      "fraudulent": "any rating <= -0.5 (raw -5) from a Benign node; negative ratings from non-benign raters ignored",
+      "seed_ids": [
+        1,
+        7,
+        35,
+        1810,
+        2642
+      ],
+      "benign_count": 636,
+      "fraudulent_count": 614,
+      "unknown_count": 4631
+    },
+    "notes": "Derived 3-class trust_label (benign/fraudulent/unknown) via the recursive Web-of-Trust recipe; evaluate classifiers on the labeled subset only, egonets use the full graph. Rated-pair structure symmetrized; rating signs/weights and timestamps not carried into edges. Label-only nodes.csv: rating aggregates would leak the derived label.",
+    "parquet": true
+  },
+  "graphs": {
+    "default": {
+      "nodes": 5881,
+      "edges": 21492,
+      "label_column": "trust_label",
+      "classes": 3,
+      "class_counts": {
+        "unknown": 4631,
+        "benign": 636,
+        "fraudulent": 614
+      },
+      "isolated_nodes": 0,
+      "feature_columns": 0
+    }
+  }
+}

blogcatalog/card.md

@@ -0,0 +1,43 @@
+# BlogCatalog (single-membership subset)
+
+**Task**: Blogger interest-group classification  
+**Size band**: medium · **Label type**: semantic  
+**Label column**: `group` · **Converter**: `converters/convert_blogcatalog.py`
+
+Blogger friendship network whose original labels are multi-label (39 overlapping interest groups; a blogger can belong to up to 11). To fit the single-label contract, the converted graph keeps only bloggers with exactly one group membership and the induced subgraph on them.
+
+## Labeling methodology (decision approved 2026-07-23)
+
+The source `group` matrix is multi-label: of 10,312 bloggers, 7,460 (72.3%) belong to exactly one group, 2,852 to two or more. Three reductions were considered — (A) keep single-membership bloggers only, (B) assign multi-member bloggers a primary group, (C) one-vs-rest on a single group. **Option A was chosen**: it is the only reduction that invents no assignment rule and keeps labels exactly as authored.
+
+- Kept: 7,460 single-membership bloggers; induced subgraph has 131,034 edges.
+- Dropped: 2,852 multi-membership bloggers and all their edges (recorded in `metadata.json` under `conversion.labeling_rule`).
+- 38 of the 39 groups survive (one group has no single-membership member).
+- Original blogger matrix indices preserved in `neext/id_mapping.csv`.
+
+Consequence: results are NOT directly comparable to multi-label BlogCatalog numbers in the embedding literature (deepwalk/node2vec macro-F1), which score all 10,312 bloggers with one-vs-rest classifiers. Implemented in `converters/convert_blogcatalog.py`.
+
+## Converted graphs (neext/)
+
+| graph | nodes | edges | classes | feature cols | isolated | class counts |
+|---|---|---|---|---|---|---|
+| default | 7,460 | 131,034 | 38 | 0 | 170 | 7: 970, 4: 597, 23: 514, 5: 504, 1: 481, 18: 432, … |
+
+*Conversion notes*: Multi-label source (39 overlapping groups). Kept only bloggers with exactly one group membership (7,460 of 10,312; 72.3%) and the induced subgraph on them; multi-member bloggers and their edges are excluded rather than force-assigned a primary group. Original blogger indices in id_mapping.csv.
+
+## Source
+
+- [blogcatalog.mat](https://raw.githubusercontent.com/phanein/deepwalk/master/example_graphs/blogcatalog.mat) — 1,255,783 bytes, sha256 `d4f4fb89ce1ccd4b…`, fetched 2026-07-23
+
+**License**: deepwalk repo GPL-3.0; data from ASU social computing repository  
+**Citation**: Tang, Liu. Relational Learning via Latent Social Dimensions. KDD 2009.
+
+- https://github.com/phanein/deepwalk
+
+## Caveats
+
+- Subset graph — 28% of bloggers (the multi-membership ones) are excluded, so published multi-label baselines don't apply.
+- 38 imbalanced classes; use macro-averaged metrics.
+
+---
+*Generated by converters/make_cards.py; stats from metadata.json. Raw files: `source/`. NEExT tables: `neext/`.*

blogcatalog/metadata.json

@@ -0,0 +1,84 @@
+{
+  "name": "blogcatalog",
+  "sources": [
+    {
+      "url": "https://raw.githubusercontent.com/phanein/deepwalk/master/example_graphs/blogcatalog.mat",
+      "final_url": "https://raw.githubusercontent.com/phanein/deepwalk/master/example_graphs/blogcatalog.mat",
+      "filename": "blogcatalog.mat",
+      "bytes": 1255783,
+      "sha256": "d4f4fb89ce1ccd4b7e2a183386c000773cc9362cc61f1be5b246a6d9c259da8f",
+      "downloaded_at": "2026-07-23T02:14:58+00:00"
+    }
+  ],
+  "title": "BlogCatalog (single-membership subset)",
+  "band": "medium",
+  "label_type": "semantic",
+  "task": "Blogger interest-group classification",
+  "license": "deepwalk repo GPL-3.0; data from ASU social computing repository",
+  "citation": "Tang, Liu. Relational Learning via Latent Social Dimensions. KDD 2009.",
+  "links": [
+    "https://github.com/phanein/deepwalk"
+  ],
+  "conversion": {
+    "script": "convert_blogcatalog.py",
+    "label_column": "group",
+    "labeling_rule": {
+      "method": "single-membership reduction (option A)",
+      "kept_nodes": 7460,
+      "dropped_multi_membership_nodes": 2852,
+      "classes_present": 38
+    },
+    "notes": "Multi-label source (39 overlapping groups). Kept only bloggers with exactly one group membership (7,460 of 10,312; 72.3%) and the induced subgraph on them; multi-member bloggers and their edges are excluded rather than force-assigned a primary group. Original blogger indices in id_mapping.csv.",
+    "parquet": true
+  },
+  "graphs": {
+    "default": {
+      "nodes": 7460,
+      "edges": 131034,
+      "label_column": "group",
+      "classes": 38,
+      "class_counts": {
+        "7": 970,
+        "4": 597,
+        "23": 514,
+        "5": 504,
+        "1": 481,
+        "18": 432,
+        "2": 352,
+        "6": 347,
+        "15": 234,
+        "10": 231,
+        "9": 221,
+        "22": 212,
+        "8": 209,
+        "19": 184,
+        "20": 182,
+        "31": 179,
+        "17": 177,
+        "25": 173,
+        "29": 173,
+        "13": 165,
+        "21": 126,
+        "3": 116,
+        "28": 91,
+        "24": 87,
+        "35": 66,
+        "0": 60,
+        "26": 58,
+        "27": 48,
+        "32": 44,
+        "30": 35,
+        "36": 34,
+        "14": 32,
+        "12": 30,
+        "34": 29,
+        "33": 26,
+        "11": 21,
+        "37": 17,
+        "38": 3
+      },
+      "isolated_nodes": 170,
+      "feature_columns": 0
+    }
+  }
+}

books/card.md

@@ -0,0 +1,31 @@
+# Books (PyGOD)
+
+**Task**: Outlier book detection (Amazon co-purchase)  
+**Size band**: small · **Label type**: anomaly  
+**Label column**: `is_outlier` · **Converter**: `converters/convert_pygod.py`
+
+Small Amazon co-purchase network of books with 21 numeric features per node and an organic binary outlier label (~2% positives). A quick, real-anomaly benchmark for the egonet outlier pipeline.
+
+## Converted graphs (neext/)
+
+| graph | nodes | edges | classes | feature cols | isolated | class counts |
+|---|---|---|---|---|---|---|
+| default | 1,418 | 3,695 | 2 | 21 | 0 | 0: 1,390, 1: 28 |
+
+*Conversion notes*: PyGOD .pt (torch pickle, trusted pygod-team/data source) via stub unpickling; edge_index symmetrized; 21 x_* feature columns.
+
+## Source
+
+- [books.pt.zip](https://raw.githubusercontent.com/pygod-team/data/main/books.pt.zip) — 60,512 bytes, sha256 `c9de331ddf4c893e…`, fetched 2026-07-23
+
+**License**: MIT (pygod-team/data)  
+**Citation**: Liu et al. BOND: Benchmarking Unsupervised Outlier Node Detection on Static Attributed Graphs. NeurIPS 2022 D&B.
+
+- https://github.com/pygod-team/data
+
+## Caveats
+
+- Source file is a torch pickle; converted via stub unpickling from the hash-verified official repo.
+
+---
+*Generated by converters/make_cards.py; stats from metadata.json. Raw files: `source/`. NEExT tables: `neext/`.*

books/metadata.json

@@ -0,0 +1,42 @@
+{
+  "name": "books",
+  "sources": [
+    {
+      "url": "https://raw.githubusercontent.com/pygod-team/data/main/books.pt.zip",
+      "final_url": "https://raw.githubusercontent.com/pygod-team/data/main/books.pt.zip",
+      "filename": "books.pt.zip",
+      "bytes": 60512,
+      "sha256": "c9de331ddf4c893e205fad552a9d1074c0fe4730edd7b47952f0c05ac360612c",
+      "downloaded_at": "2026-07-23T02:14:16+00:00"
+    }
+  ],
+  "conversion": {
+    "script": "convert_pygod.py",
+    "label_column": "is_outlier",
+    "notes": "PyGOD .pt (torch pickle, trusted pygod-team/data source) via stub unpickling; edge_index symmetrized; 21 x_* feature columns.",
+    "parquet": true
+  },
+  "graphs": {
+    "default": {
+      "nodes": 1418,
+      "edges": 3695,
+      "label_column": "is_outlier",
+      "classes": 2,
+      "class_counts": {
+        "0": 1390,
+        "1": 28
+      },
+      "isolated_nodes": 0,
+      "feature_columns": 21
+    }
+  },
+  "title": "Books (PyGOD)",
+  "band": "small",
+  "label_type": "anomaly",
+  "task": "Outlier book detection (Amazon co-purchase)",
+  "license": "MIT (pygod-team/data)",
+  "citation": "Liu et al. BOND: Benchmarking Unsupervised Outlier Node Detection on Static Attributed Graphs. NeurIPS 2022 D&B.",
+  "links": [
+    "https://github.com/pygod-team/data"
+  ]
+}

citeseer/card.md

@@ -0,0 +1,31 @@
+# Citeseer
+
+**Task**: Paper topic classification (citation network)  
+**Size band**: small · **Label type**: semantic  
+**Label column**: `subject` · **Converter**: `converters/convert_linqs.py`
+
+Citation network companion to Cora with 6 subject classes and 3,703 binary bag-of-words features. String paper IDs were remapped to integers (see id_mapping.csv).
+
+## Converted graphs (neext/)
+
+| graph | nodes | edges | classes | feature cols | isolated | class counts |
+|---|---|---|---|---|---|---|
+| default | 3,312 | 4,536 | 6 | 3703 | 48 | DB: 701, IR: 668, Agents: 596, ML: 590, HCI: 508, AI: 249 |
+
+*Conversion notes*: LINQS .content/.cites; 3703 binary bag-of-words feature columns; 17 citation edges dropped (endpoint not in .content); source paper IDs remapped (id_mapping.csv).
+
+## Source
+
+- [citeseer.tgz](https://linqs-data.soe.ucsc.edu/public/lbc/citeseer.tgz) — 359,425 bytes, sha256 `b02ee7b5d83130f8…`, fetched 2026-07-23
+
+**License**: LINQS research distribution  
+**Citation**: Sen, Namata, Bilgic, Getoor, Gallagher, Eliassi-Rad. Collective Classification in Network Data. AI Magazine 2008.
+
+- https://linqs.org/datasets/
+
+## Caveats
+
+- 17 citation edges reference papers absent from the content file and were dropped (recorded in metadata).
+
+---
+*Generated by converters/make_cards.py; stats from metadata.json. Raw files: `source/`. NEExT tables: `neext/`.*

citeseer/metadata.json

@@ -0,0 +1,47 @@
+{
+  "name": "citeseer",
+  "sources": [
+    {
+      "url": "https://linqs-data.soe.ucsc.edu/public/lbc/citeseer.tgz",
+      "final_url": "https://linqs-data.soe.ucsc.edu/public/lbc/citeseer.tgz",
+      "filename": "citeseer.tgz",
+      "bytes": 359425,
+      "sha256": "b02ee7b5d83130f8fd45b59017a76fdae3e998629a1904c2c5e07343a9664685",
+      "downloaded_at": "2026-07-23T02:14:16+00:00"
+    }
+  ],
+  "conversion": {
+    "script": "convert_linqs.py",
+    "label_column": "subject",
+    "dropped_edges_unknown_endpoint": 17,
+    "notes": "LINQS .content/.cites; 3703 binary bag-of-words feature columns; 17 citation edges dropped (endpoint not in .content); source paper IDs remapped (id_mapping.csv).",
+    "parquet": true
+  },
+  "graphs": {
+    "default": {
+      "nodes": 3312,
+      "edges": 4536,
+      "label_column": "subject",
+      "classes": 6,
+      "class_counts": {
+        "DB": 701,
+        "IR": 668,
+        "Agents": 596,
+        "ML": 590,
+        "HCI": 508,
+        "AI": 249
+      },
+      "isolated_nodes": 48,
+      "feature_columns": 3703
+    }
+  },
+  "title": "Citeseer",
+  "band": "small",
+  "label_type": "semantic",
+  "task": "Paper topic classification (citation network)",
+  "license": "LINQS research distribution",
+  "citation": "Sen, Namata, Bilgic, Getoor, Gallagher, Eliassi-Rad. Collective Classification in Network Data. AI Magazine 2008.",
+  "links": [
+    "https://linqs.org/datasets/"
+  ]
+}

coauthor-cs/card.md

@@ -0,0 +1,31 @@
+# Coauthor CS
+
+**Task**: Research-field classification (co-authorship)  
+**Size band**: medium · **Label type**: semantic  
+**Label column**: `field` · **Converter**: `converters/convert_npz.py`
+
+Microsoft Academic co-authorship graph in computer science; the label is the author's field (15 classes).
+
+## Converted graphs (neext/)
+
+| graph | nodes | edges | classes | feature cols | isolated | class counts |
+|---|---|---|---|---|---|---|
+| default | 18,333 | 81,894 | 15 | 0 | 0 | 13: 4,136, 5: 2,193, 2: 2,050, 11: 2,033, 10: 1,444, 4: 1,394, … |
+
+*Conversion notes*: shchur gnn-benchmark npz; CSR adjacency symmetrized. Feature matrix too large for CSV (skipped; available in source npz).
+
+## Source
+
+- [ms_academic_cs.npz](https://raw.githubusercontent.com/shchur/gnn-benchmark/master/data/npz/ms_academic_cs.npz) — 12,835,626 bytes, sha256 `933c745734e78908…`, fetched 2026-07-23
+
+**License**: MIT (shchur packaging)  
+**Citation**: Shchur et al. Pitfalls of GNN Evaluation. 2018.
+
+- https://github.com/shchur/gnn-benchmark
+
+## Caveats
+
+- 6,805-dim keyword features exceed the CSV budget — label-only nodes.csv; features remain in the source npz.
+
+---
+*Generated by converters/make_cards.py; stats from metadata.json. Raw files: `source/`. NEExT tables: `neext/`.*

coauthor-cs/metadata.json

@@ -0,0 +1,56 @@
+{
+  "name": "coauthor-cs",
+  "sources": [
+    {
+      "url": "https://raw.githubusercontent.com/shchur/gnn-benchmark/master/data/npz/ms_academic_cs.npz",
+      "final_url": "https://raw.githubusercontent.com/shchur/gnn-benchmark/master/data/npz/ms_academic_cs.npz",
+      "filename": "ms_academic_cs.npz",
+      "bytes": 12835626,
+      "sha256": "933c745734e78908c7eb77172a673fee0503f30e26e5a99b36814d88534c42e3",
+      "downloaded_at": "2026-07-23T02:14:49+00:00"
+    }
+  ],
+  "conversion": {
+    "script": "convert_npz.py",
+    "label_column": "field",
+    "notes": "shchur gnn-benchmark npz; CSR adjacency symmetrized. Feature matrix too large for CSV (skipped; available in source npz).",
+    "parquet": true,
+    "parquet_notes": "nodes.parquet includes the full feature matrix omitted from nodes.csv for size."
+  },
+  "graphs": {
+    "default": {
+      "nodes": 18333,
+      "edges": 81894,
+      "label_column": "field",
+      "classes": 15,
+      "class_counts": {
+        "13": 4136,
+        "5": 2193,
+        "2": 2050,
+        "11": 2033,
+        "10": 1444,
+        "4": 1394,
+        "7": 924,
+        "14": 876,
+        "8": 775,
+        "0": 708,
+        "1": 462,
+        "3": 429,
+        "12": 420,
+        "6": 371,
+        "9": 118
+      },
+      "isolated_nodes": 0,
+      "feature_columns": 0
+    }
+  },
+  "title": "Coauthor CS",
+  "band": "medium",
+  "label_type": "semantic",
+  "task": "Research-field classification (co-authorship)",
+  "license": "MIT (shchur packaging)",
+  "citation": "Shchur et al. Pitfalls of GNN Evaluation. 2018.",
+  "links": [
+    "https://github.com/shchur/gnn-benchmark"
+  ]
+}

coauthor-physics/card.md

@@ -0,0 +1,31 @@
+# Coauthor Physics
+
+**Task**: Research-field classification (co-authorship)  
+**Size band**: medium · **Label type**: semantic  
+**Label column**: `field` · **Converter**: `converters/convert_npz.py`
+
+Microsoft Academic co-authorship graph in physics; 5 field classes; the largest shchur benchmark (34.5k nodes).
+
+## Converted graphs (neext/)
+
+| graph | nodes | edges | classes | feature cols | isolated | class counts |
+|---|---|---|---|---|---|---|
+| default | 34,493 | 247,962 | 5 | 0 | 0 | 2: 17,426, 0: 5,750, 1: 5,045, 4: 3,519, 3: 2,753 |
+
+*Conversion notes*: shchur gnn-benchmark npz; CSR adjacency symmetrized. Feature matrix too large for CSV (skipped; available in source npz).
+
+## Source
+
+- [ms_academic_phy.npz](https://raw.githubusercontent.com/shchur/gnn-benchmark/master/data/npz/ms_academic_phy.npz) — 16,374,005 bytes, sha256 `e4d68468eba5fb8f…`, fetched 2026-07-23
+
+**License**: MIT (shchur packaging)  
+**Citation**: Shchur et al. Pitfalls of GNN Evaluation. 2018.
+
+- https://github.com/shchur/gnn-benchmark
+
+## Caveats
+
+- 8,415-dim features exceed the CSV budget — label-only nodes.csv; features remain in the source npz.
+
+---
+*Generated by converters/make_cards.py; stats from metadata.json. Raw files: `source/`. NEExT tables: `neext/`.*

coauthor-physics/metadata.json

@@ -0,0 +1,46 @@
+{
+  "name": "coauthor-physics",
+  "sources": [
+    {
+      "url": "https://raw.githubusercontent.com/shchur/gnn-benchmark/master/data/npz/ms_academic_phy.npz",
+      "final_url": "https://raw.githubusercontent.com/shchur/gnn-benchmark/master/data/npz/ms_academic_phy.npz",
+      "filename": "ms_academic_phy.npz",
+      "bytes": 16374005,
+      "sha256": "e4d68468eba5fb8f6b280ffd406bd53e6e4295ba1dd2967d6fe99b978e4b3375",
+      "downloaded_at": "2026-07-23T02:14:50+00:00"
+    }
+  ],
+  "conversion": {
+    "script": "convert_npz.py",
+    "label_column": "field",
+    "notes": "shchur gnn-benchmark npz; CSR adjacency symmetrized. Feature matrix too large for CSV (skipped; available in source npz).",
+    "parquet": true,
+    "parquet_notes": "nodes.parquet includes the full feature matrix omitted from nodes.csv for size."
+  },
+  "graphs": {
+    "default": {
+      "nodes": 34493,
+      "edges": 247962,
+      "label_column": "field",
+      "classes": 5,
+      "class_counts": {
+        "2": 17426,
+        "0": 5750,
+        "1": 5045,
+        "4": 3519,
+        "3": 2753
+      },
+      "isolated_nodes": 0,
+      "feature_columns": 0
+    }
+  },
+  "title": "Coauthor Physics",
+  "band": "medium",
+  "label_type": "semantic",
+  "task": "Research-field classification (co-authorship)",
+  "license": "MIT (shchur packaging)",
+  "citation": "Shchur et al. Pitfalls of GNN Evaluation. 2018.",
+  "links": [
+    "https://github.com/shchur/gnn-benchmark"
+  ]
+}

cora/card.md

@@ -0,0 +1,32 @@
+# Cora
+
+**Task**: Paper topic classification (citation network)  
+**Size band**: small · **Label type**: semantic  
+**Label column**: `subject` · **Converter**: `converters/convert_linqs.py`
+
+Classic citation network of machine-learning papers; the label is one of 7 subject areas and features are 1,433 binary bag-of-words indicators. Strongly homophilous — useful as a calibration baseline against the GNN literature.
+
+## Converted graphs (neext/)
+
+| graph | nodes | edges | classes | feature cols | isolated | class counts |
+|---|---|---|---|---|---|---|
+| default | 2,708 | 5,278 | 7 | 1433 | 0 | Neural_Networks: 818, Probabilistic_Methods: 426, Genetic_Algorithms: 418, Theory: 351, Case_Based: 298, Reinforcement_Learning: 217, … |
+
+*Conversion notes*: LINQS .content/.cites; 1433 binary bag-of-words feature columns; 0 citation edges dropped (endpoint not in .content); source paper IDs remapped (id_mapping.csv).
+
+## Source
+
+- [cora.tgz](https://linqs-data.soe.ucsc.edu/public/lbc/cora.tgz) — 168,052 bytes, sha256 `0d4ed463d1627bb7…`, fetched 2026-07-23
+
+**License**: LINQS research distribution  
+**Citation**: Sen, Namata, Bilgic, Getoor, Gallagher, Eliassi-Rad. Collective Classification in Network Data. AI Magazine 2008.
+
+- https://linqs.org/datasets/
+- https://github.com/kimiyoung/planetoid
+
+## Caveats
+
+- Homophily-driven labels favor message-passing GNNs; egonet embeddings are expected to trail here.
+
+---
+*Generated by converters/make_cards.py; stats from metadata.json. Raw files: `source/`. NEExT tables: `neext/`.*

cora/metadata.json

@@ -0,0 +1,49 @@
+{
+  "name": "cora",
+  "sources": [
+    {
+      "url": "https://linqs-data.soe.ucsc.edu/public/lbc/cora.tgz",
+      "final_url": "https://linqs-data.soe.ucsc.edu/public/lbc/cora.tgz",
+      "filename": "cora.tgz",
+      "bytes": 168052,
+      "sha256": "0d4ed463d1627bb7f3e8420effe8f5545fd492ae8f88dab44ce86cee7b26d7e8",
+      "downloaded_at": "2026-07-23T02:14:15+00:00"
+    }
+  ],
+  "conversion": {
+    "script": "convert_linqs.py",
+    "label_column": "subject",
+    "dropped_edges_unknown_endpoint": 0,
+    "notes": "LINQS .content/.cites; 1433 binary bag-of-words feature columns; 0 citation edges dropped (endpoint not in .content); source paper IDs remapped (id_mapping.csv).",
+    "parquet": true
+  },
+  "graphs": {
+    "default": {
+      "nodes": 2708,
+      "edges": 5278,
+      "label_column": "subject",
+      "classes": 7,
+      "class_counts": {
+        "Neural_Networks": 818,
+        "Probabilistic_Methods": 426,
+        "Genetic_Algorithms": 418,
+        "Theory": 351,
+        "Case_Based": 298,
+        "Reinforcement_Learning": 217,
+        "Rule_Learning": 180
+      },
+      "isolated_nodes": 0,
+      "feature_columns": 1433
+    }
+  },
+  "title": "Cora",
+  "band": "small",
+  "label_type": "semantic",
+  "task": "Paper topic classification (citation network)",
+  "license": "LINQS research distribution",
+  "citation": "Sen, Namata, Bilgic, Getoor, Gallagher, Eliassi-Rad. Collective Classification in Network Data. AI Magazine 2008.",
+  "links": [
+    "https://linqs.org/datasets/",
+    "https://github.com/kimiyoung/planetoid"
+  ]
+}

deezer-europe/card.md

@@ -0,0 +1,31 @@
+# Deezer Europe
+
+**Task**: User gender classification  
+**Size band**: medium · **Label type**: semantic  
+**Label column**: `gender` · **Converter**: `converters/convert_musae.py`
+
+Social network of European Deezer users with a binary gender label. A near-random-baseline-hard social prediction task; useful as a difficulty contrast to more structural targets.
+
+## Converted graphs (neext/)
+
+| graph | nodes | edges | classes | feature cols | isolated | class counts |
+|---|---|---|---|---|---|---|
+| default | 28,281 | 92,752 | 2 | 0 | 0 | 0: 15,743, 1: 12,538 |
+
+*Conversion notes*: Label = binary gender. Feature JSON (liked artists) not columnized.
+
+## Source
+
+- [deezer_europe.zip](https://snap.stanford.edu/data/deezer_europe.zip) — 2,622,306 bytes, sha256 `dd66a73f8d8690b5…`, fetched 2026-07-23
+
+**License**: SNAP; cite FEATHER  
+**Citation**: Rozemberczki, Sarkar. FEATHER. CIKM 2020.
+
+- https://snap.stanford.edu/data/feather-deezer-social.html
+
+## Caveats
+
+- Feature JSON not columnized.
+
+---
+*Generated by converters/make_cards.py; stats from metadata.json. Raw files: `source/`. NEExT tables: `neext/`.*

deezer-europe/metadata.json

@@ -0,0 +1,42 @@
+{
+  "name": "deezer-europe",
+  "sources": [
+    {
+      "url": "https://snap.stanford.edu/data/deezer_europe.zip",
+      "final_url": "https://snap.stanford.edu/data/deezer_europe.zip",
+      "filename": "deezer_europe.zip",
+      "bytes": 2622306,
+      "sha256": "dd66a73f8d8690b5bc300ba378883fb2c2f6316aec8917b6a2428e352fc9e498",
+      "downloaded_at": "2026-07-23T02:14:26+00:00"
+    }
+  ],
+  "conversion": {
+    "script": "convert_musae.py",
+    "label_column": "gender",
+    "notes": "Label = binary gender. Feature JSON (liked artists) not columnized.",
+    "parquet": true
+  },
+  "graphs": {
+    "default": {
+      "nodes": 28281,
+      "edges": 92752,
+      "label_column": "gender",
+      "classes": 2,
+      "class_counts": {
+        "0": 15743,
+        "1": 12538
+      },
+      "isolated_nodes": 0,
+      "feature_columns": 0
+    }
+  },
+  "title": "Deezer Europe",
+  "band": "medium",
+  "label_type": "semantic",
+  "task": "User gender classification",
+  "license": "SNAP; cite FEATHER",
+  "citation": "Rozemberczki, Sarkar. FEATHER. CIKM 2020.",
+  "links": [
+    "https://snap.stanford.edu/data/feather-deezer-social.html"
+  ]
+}

email-eu-core/card.md

@@ -0,0 +1,32 @@
+# Email-EU-core
+
+**Task**: Department classification from email traffic  
+**Size band**: small · **Label type**: structural  
+**Label column**: `department` · **Converter**: `converters/convert_small_text.py`
+
+Internal email network of a European research institution; an edge means at least one email between two members. The label is the member's department (42 classes). No node features exist, so classification is a structure-only problem.
+
+## Converted graphs (neext/)
+
+| graph | nodes | edges | classes | feature cols | isolated | class counts |
+|---|---|---|---|---|---|---|
+| default | 1,005 | 16,064 | 42 | 0 | 19 | 4: 109, 14: 92, 1: 65, 21: 61, 15: 55, 7: 51, … |
+
+*Conversion notes*: Directed email edges symmetrized/deduplicated; department labels 0-41; no node features (structure-only).
+
+## Source
+
+- [email-Eu-core.txt.gz](https://snap.stanford.edu/data/email-Eu-core.txt.gz) — 79,754 bytes, sha256 `4b47acdb80197b08…`, fetched 2026-07-23
+- [email-Eu-core-department-labels.txt.gz](https://snap.stanford.edu/data/email-Eu-core-department-labels.txt.gz) — 2,663 bytes, sha256 `e5abe5b4581a4800…`, fetched 2026-07-23
+
+**License**: SNAP research use; cite the source papers  
+**Citation**: Yin, Benson, Leskovec, Gleich. Local Higher-Order Graph Clustering. KDD 2017.
+
+- https://snap.stanford.edu/data/email-Eu-core.html
+
+## Caveats
+
+- 42 classes over 1,005 nodes — many tiny departments; expect a hard task and consider grouping small classes.
+
+---
+*Generated by converters/make_cards.py; stats from metadata.json. Raw files: `source/`. NEExT tables: `neext/`.*

email-eu-core/metadata.json

@@ -0,0 +1,90 @@
+{
+  "name": "email-eu-core",
+  "sources": [
+    {
+      "url": "https://snap.stanford.edu/data/email-Eu-core.txt.gz",
+      "final_url": "https://snap.stanford.edu/data/email-Eu-core.txt.gz",
+      "filename": "email-Eu-core.txt.gz",
+      "bytes": 79754,
+      "sha256": "4b47acdb80197b085fe63c819c357ae488131ee904ed93d1b219a68b0f9e245f",
+      "downloaded_at": "2026-07-23T02:14:13+00:00"
+    },
+    {
+      "url": "https://snap.stanford.edu/data/email-Eu-core-department-labels.txt.gz",
+      "final_url": "https://snap.stanford.edu/data/email-Eu-core-department-labels.txt.gz",
+      "filename": "email-Eu-core-department-labels.txt.gz",
+      "bytes": 2663,
+      "sha256": "e5abe5b4581a480032a63adcf2576c161785f45692642c6ebb0b1276f0f33669",
+      "downloaded_at": "2026-07-23T02:14:13+00:00"
+    }
+  ],
+  "conversion": {
+    "script": "convert_small_text.py",
+    "label_column": "department",
+    "notes": "Directed email edges symmetrized/deduplicated; department labels 0-41; no node features (structure-only).",
+    "parquet": true
+  },
+  "graphs": {
+    "default": {
+      "nodes": 1005,
+      "edges": 16064,
+      "label_column": "department",
+      "classes": 42,
+      "class_counts": {
+        "4": 109,
+        "14": 92,
+        "1": 65,
+        "21": 61,
+        "15": 55,
+        "7": 51,
+        "0": 49,
+        "10": 39,
+        "17": 35,
+        "9": 32,
+        "19": 29,
+        "11": 29,
+        "6": 28,
+        "23": 27,
+        "13": 26,
+        "16": 25,
+        "22": 25,
+        "36": 22,
+        "8": 19,
+        "5": 18,
+        "37": 15,
+        "20": 14,
+        "35": 13,
+        "34": 13,
+        "38": 13,
+        "3": 12,
+        "27": 10,
+        "2": 10,
+        "26": 9,
+        "32": 9,
+        "28": 8,
+        "31": 8,
+        "25": 6,
+        "24": 6,
+        "29": 5,
+        "30": 4,
+        "40": 4,
+        "12": 3,
+        "39": 3,
+        "41": 2,
+        "18": 1,
+        "33": 1
+      },
+      "isolated_nodes": 19,
+      "feature_columns": 0
+    }
+  },
+  "title": "Email-EU-core",
+  "band": "small",
+  "label_type": "structural",
+  "task": "Department classification from email traffic",
+  "license": "SNAP research use; cite the source papers",
+  "citation": "Yin, Benson, Leskovec, Gleich. Local Higher-Order Graph Clustering. KDD 2017.",
+  "links": [
+    "https://snap.stanford.edu/data/email-Eu-core.html"
+  ]
+}

facebook-page-page/card.md

@@ -0,0 +1,33 @@
+# Facebook Page-Page
+
+**Task**: Page category classification  
+**Size band**: medium · **Label type**: semantic  
+**Label column**: `page_type` · **Converter**: `converters/convert_musae.py`
+
+Network of verified Facebook pages with mutual-like edges; the label is the page category (politician / governmental / tvshow / company). Standard MUSAE benchmark.
+
+## Converted graphs (neext/)
+
+| graph | nodes | edges | classes | feature cols | isolated | class counts |
+|---|---|---|---|---|---|---|
+| default | 22,470 | 170,823 | 4 | 0 | 0 | government: 6,880, company: 6,495, politician: 5,768, tvshow: 3,327 |
+
+*Conversion notes*: Label = page_type (politician/governmental/tvshow/company). Page names and facebook_ids dropped; feature JSON not columnized.
+
+## Source
+
+- [facebook_edges.csv](https://raw.githubusercontent.com/benedekrozemberczki/MUSAE/master/input/edges/facebook_edges.csv) — 1,882,610 bytes, sha256 `7c50d8f02a75cc08…`, fetched 2026-07-23
+- [facebook_target.csv](https://raw.githubusercontent.com/benedekrozemberczki/MUSAE/master/input/target/facebook_target.csv) — 1,177,912 bytes, sha256 `7bd96eafea3c2ca4…`, fetched 2026-07-23
+- [facebook.json](https://raw.githubusercontent.com/benedekrozemberczki/MUSAE/master/input/features/facebook.json) — 2,088,766 bytes, sha256 `ea870537646a9364…`, fetched 2026-07-23
+
+**License**: MUSAE repo GPL-3.0; cite MUSAE  
+**Citation**: Rozemberczki, Allen, Sarkar. MUSAE. 2021.
+
+- https://github.com/benedekrozemberczki/MUSAE
+
+## Caveats
+
+- SNAP's facebook_large.zip is 404 — the MUSAE GitHub mirror is the working source (recorded here).
+
+---
+*Generated by converters/make_cards.py; stats from metadata.json. Raw files: `source/`. NEExT tables: `neext/`.*

facebook-page-page/metadata.json

@@ -0,0 +1,60 @@
+{
+  "name": "facebook-page-page",
+  "sources": [
+    {
+      "url": "https://raw.githubusercontent.com/benedekrozemberczki/MUSAE/master/input/edges/facebook_edges.csv",
+      "final_url": "https://raw.githubusercontent.com/benedekrozemberczki/MUSAE/master/input/edges/facebook_edges.csv",
+      "filename": "facebook_edges.csv",
+      "bytes": 1882610,
+      "sha256": "7c50d8f02a75cc0829577814a1fc14535164daa38d79c3612340c9e9cdbd4022",
+      "downloaded_at": "2026-07-23T02:14:29+00:00"
+    },
+    {
+      "url": "https://raw.githubusercontent.com/benedekrozemberczki/MUSAE/master/input/target/facebook_target.csv",
+      "final_url": "https://raw.githubusercontent.com/benedekrozemberczki/MUSAE/master/input/target/facebook_target.csv",
+      "filename": "facebook_target.csv",
+      "bytes": 1177912,
+      "sha256": "7bd96eafea3c2ca40f44bfa9e73642194696c21e38c7b29ed409c32ad14075cd",
+      "downloaded_at": "2026-07-23T02:14:29+00:00"
+    },
+    {
+      "url": "https://raw.githubusercontent.com/benedekrozemberczki/MUSAE/master/input/features/facebook.json",
+      "final_url": "https://raw.githubusercontent.com/benedekrozemberczki/MUSAE/master/input/features/facebook.json",
+      "filename": "facebook.json",
+      "bytes": 2088766,
+      "sha256": "ea870537646a93642a0008d38aa9bfeff02018070ba97b1f3f469d9622626436",
+      "downloaded_at": "2026-07-23T02:14:29+00:00"
+    }
+  ],
+  "conversion": {
+    "script": "convert_musae.py",
+    "label_column": "page_type",
+    "notes": "Label = page_type (politician/governmental/tvshow/company). Page names and facebook_ids dropped; feature JSON not columnized.",
+    "parquet": true
+  },
+  "graphs": {
+    "default": {
+      "nodes": 22470,
+      "edges": 170823,
+      "label_column": "page_type",
+      "classes": 4,
+      "class_counts": {
+        "government": 6880,
+        "company": 6495,
+        "politician": 5768,
+        "tvshow": 3327
+      },
+      "isolated_nodes": 0,
+      "feature_columns": 0
+    }
+  },
+  "title": "Facebook Page-Page",
+  "band": "medium",
+  "label_type": "semantic",
+  "task": "Page category classification",
+  "license": "MUSAE repo GPL-3.0; cite MUSAE",
+  "citation": "Rozemberczki, Allen, Sarkar. MUSAE. 2021.",
+  "links": [
+    "https://github.com/benedekrozemberczki/MUSAE"
+  ]
+}

lastfm-asia/card.md

@@ -0,0 +1,31 @@
+# LastFM Asia
+
+**Task**: User country classification  
+**Size band**: medium · **Label type**: semantic  
+**Label column**: `country` · **Converter**: `converters/convert_musae.py`
+
+Social network of LastFM users in Asia; the label is the user's country (18 classes, imbalanced). Clean integer-ID CSVs from the FEATHER release.
+
+## Converted graphs (neext/)
+
+| graph | nodes | edges | classes | feature cols | isolated | class counts |
+|---|---|---|---|---|---|---|
+| default | 7,624 | 27,806 | 18 | 0 | 0 | 17: 1,572, 10: 1,303, 0: 1,098, 6: 655, 14: 570, 3: 515, … |
+
+*Conversion notes*: Label = country id (18 classes). Feature JSON (liked artists) not columnized.
+
+## Source
+
+- [lastfm_asia.zip](https://snap.stanford.edu/data/lastfm_asia.zip) — 6,527,202 bytes, sha256 `51acb78a923bb223…`, fetched 2026-07-23
+
+**License**: SNAP; cite FEATHER  
+**Citation**: Rozemberczki, Sarkar. Characteristic Functions on Graphs (FEATHER). CIKM 2020.
+
+- https://snap.stanford.edu/data/feather-lastfm-social.html
+
+## Caveats
+
+- Feature JSON (liked artists) not columnized.
+
+---
+*Generated by converters/make_cards.py; stats from metadata.json. Raw files: `source/`. NEExT tables: `neext/`.*

lastfm-asia/metadata.json

@@ -0,0 +1,58 @@
+{
+  "name": "lastfm-asia",
+  "sources": [
+    {
+      "url": "https://snap.stanford.edu/data/lastfm_asia.zip",
+      "final_url": "https://snap.stanford.edu/data/lastfm_asia.zip",
+      "filename": "lastfm_asia.zip",
+      "bytes": 6527202,
+      "sha256": "51acb78a923bb223ed6e61be88f91122fb29adca3f07beff7289cafd98601d47",
+      "downloaded_at": "2026-07-23T02:14:23+00:00"
+    }
+  ],
+  "conversion": {
+    "script": "convert_musae.py",
+    "label_column": "country",
+    "notes": "Label = country id (18 classes). Feature JSON (liked artists) not columnized.",
+    "parquet": true
+  },
+  "graphs": {
+    "default": {
+      "nodes": 7624,
+      "edges": 27806,
+      "label_column": "country",
+      "classes": 18,
+      "class_counts": {
+        "17": 1572,
+        "10": 1303,
+        "0": 1098,
+        "6": 655,
+        "14": 570,
+        "3": 515,
+        "8": 468,
+        "5": 391,
+        "15": 257,
+        "16": 254,
+        "11": 138,
+        "7": 82,
+        "2": 73,
+        "13": 63,
+        "9": 58,
+        "12": 57,
+        "1": 54,
+        "4": 16
+      },
+      "isolated_nodes": 0,
+      "feature_columns": 0
+    }
+  },
+  "title": "LastFM Asia",
+  "band": "medium",
+  "label_type": "semantic",
+  "task": "User country classification",
+  "license": "SNAP; cite FEATHER",
+  "citation": "Rozemberczki, Sarkar. Characteristic Functions on Graphs (FEATHER). CIKM 2020.",
+  "links": [
+    "https://snap.stanford.edu/data/feather-lastfm-social.html"
+  ]
+}

minesweeper/card.md

@@ -0,0 +1,31 @@
+# Minesweeper
+
+**Task**: Mine prediction on a synthetic grid  
+**Size band**: medium · **Label type**: structural  
+**Label column**: `is_mine` · **Converter**: `converters/convert_npz.py`
+
+Synthetic 100x100 grid graph in the minesweeper style: 20% of cells are mines, features are one-hot counts of neighboring mines. Fully structural and regular — a controlled probe of what egonet embeddings can and cannot see.
+
+## Converted graphs (neext/)
+
+| graph | nodes | edges | classes | feature cols | isolated | class counts |
+|---|---|---|---|---|---|---|
+| default | 10,000 | 39,402 | 2 | 7 | 0 | 0: 8,000, 1: 2,000 |
+
+*Conversion notes*: yandex heterophilous-graphs npz; edges symmetrized; 7 f_* feature columns.
+
+## Source
+
+- [minesweeper.npz](https://raw.githubusercontent.com/yandex-research/heterophilous-graphs/main/data/minesweeper.npz) — 135,045 bytes, sha256 `e664c8dacf1e8ac4…`, fetched 2026-07-23
+
+**License**: MIT (yandex-research)  
+**Citation**: Platonov et al. ICLR 2023.
+
+- https://github.com/yandex-research/heterophilous-graphs
+
+## Caveats
+
+- Synthetic; the known optimal strategy bounds achievable AUC.
+
+---
+*Generated by converters/make_cards.py; stats from metadata.json. Raw files: `source/`. NEExT tables: `neext/`.*

minesweeper/metadata.json

@@ -0,0 +1,42 @@
+{
+  "name": "minesweeper",
+  "sources": [
+    {
+      "url": "https://raw.githubusercontent.com/yandex-research/heterophilous-graphs/main/data/minesweeper.npz",
+      "final_url": "https://raw.githubusercontent.com/yandex-research/heterophilous-graphs/main/data/minesweeper.npz",
+      "filename": "minesweeper.npz",
+      "bytes": 135045,
+      "sha256": "e664c8dacf1e8ac466c2c09ed4b237bd2c5541f47a6eae9c6092cb87f16412b3",
+      "downloaded_at": "2026-07-23T02:14:53+00:00"
+    }
+  ],
+  "conversion": {
+    "script": "convert_npz.py",
+    "label_column": "is_mine",
+    "notes": "yandex heterophilous-graphs npz; edges symmetrized; 7 f_* feature columns.",
+    "parquet": true
+  },
+  "graphs": {
+    "default": {
+      "nodes": 10000,
+      "edges": 39402,
+      "label_column": "is_mine",
+      "classes": 2,
+      "class_counts": {
+        "0": 8000,
+        "1": 2000
+      },
+      "isolated_nodes": 0,
+      "feature_columns": 7
+    }
+  },
+  "title": "Minesweeper",
+  "band": "medium",
+  "label_type": "structural",
+  "task": "Mine prediction on a synthetic grid",
+  "license": "MIT (yandex-research)",
+  "citation": "Platonov et al. ICLR 2023.",
+  "links": [
+    "https://github.com/yandex-research/heterophilous-graphs"
+  ]
+}

ogbn-arxiv/card.md

@@ -0,0 +1,31 @@
+# ogbn-arxiv
+
+**Task**: arXiv subject-area classification  
+**Size band**: large · **Label type**: semantic  
+**Label column**: `arxiv_category` · **Converter**: `converters/convert_large.py`
+
+Citation network of all arXiv CS papers from MAG; 40 subject classes mapped to readable names (e.g. 'arxiv cs lg'), 128-dim word2vec features. The standard large-scale leaderboard benchmark.
+
+## Converted graphs (neext/)
+
+| graph | nodes | edges | classes | feature cols | isolated | class counts |
+|---|---|---|---|---|---|---|
+| default | 169,343 | 1,157,799 | 40 | 128 | 0 | arxiv cs cv: 27,321, arxiv cs lg: 22,187, arxiv cs it: 21,406, arxiv cs cl: 11,814, arxiv cs ai: 7,869, arxiv cs ds: 7,867, … |
+
+*Conversion notes*: OGB raw CSVs; 40 arXiv subject classes mapped to readable names; 128 word2vec emb_* columns (rounded); directed citations symmetrized.
+
+## Source
+
+- [arxiv.zip](https://snap.stanford.edu/ogb/data/nodeproppred/arxiv.zip) — 83,058,288 bytes, sha256 `49f85c801589ecdc…`, fetched 2026-07-23
+
+**License**: ODC-BY  
+**Citation**: Hu, Fey, Zitnik, Dong, Ren, Liu, Catasta, Leskovec. Open Graph Benchmark. NeurIPS 2020.
+
+- https://ogb.stanford.edu/docs/nodeprop/#ogbn-arxiv
+
+## Caveats
+
+- 169k nodes -> 169k egonets un-sampled; core-library scripts with sample_fraction, not Workbench browsing.
+
+---
+*Generated by converters/make_cards.py; stats from metadata.json. Raw files: `source/`. NEExT tables: `neext/`.*

ogbn-arxiv/metadata.json

@@ -0,0 +1,80 @@
+{
+  "name": "ogbn-arxiv",
+  "sources": [
+    {
+      "url": "https://snap.stanford.edu/ogb/data/nodeproppred/arxiv.zip",
+      "final_url": "https://snap.stanford.edu/ogb/data/nodeproppred/arxiv.zip",
+      "filename": "arxiv.zip",
+      "bytes": 83058288,
+      "sha256": "49f85c801589ecdcc52cfaca99693aaea7b8af16a9ac3f41dd85a5f3193fe276",
+      "downloaded_at": "2026-07-23T02:15:04+00:00"
+    }
+  ],
+  "conversion": {
+    "script": "convert_large.py",
+    "label_column": "arxiv_category",
+    "notes": "OGB raw CSVs; 40 arXiv subject classes mapped to readable names; 128 word2vec emb_* columns (rounded); directed citations symmetrized.",
+    "parquet": true
+  },
+  "graphs": {
+    "default": {
+      "nodes": 169343,
+      "edges": 1157799,
+      "label_column": "arxiv_category",
+      "classes": 40,
+      "class_counts": {
+        "arxiv cs cv": 27321,
+        "arxiv cs lg": 22187,
+        "arxiv cs it": 21406,
+        "arxiv cs cl": 11814,
+        "arxiv cs ai": 7869,
+        "arxiv cs ds": 7867,
+        "arxiv cs ni": 6232,
+        "arxiv cs cr": 5862,
+        "arxiv cs dc": 4958,
+        "arxiv cs lo": 4839,
+        "arxiv cs ro": 4801,
+        "arxiv cs si": 4605,
+        "arxiv cs gt": 3524,
+        "arxiv cs sy": 2877,
+        "arxiv cs se": 2834,
+        "arxiv cs ir": 2828,
+        "arxiv cs cc": 2820,
+        "arxiv cs db": 2369,
+        "arxiv cs ne": 2358,
+        "arxiv cs cy": 2080,
+        "arxiv cs cg": 2076,
+        "arxiv cs dm": 2029,
+        "arxiv cs pl": 1903,
+        "arxiv cs hc": 1618,
+        "arxiv cs dl": 1507,
+        "arxiv cs fl": 1271,
+        "arxiv cs sd": 1257,
+        "arxiv cs ma": 750,
+        "arxiv cs et": 749,
+        "arxiv cs mm": 687,
+        "arxiv cs sc": 597,
+        "arxiv cs ce": 589,
+        "arxiv cs na": 565,
+        "arxiv cs gr": 515,
+        "arxiv cs pf": 416,
+        "arxiv cs ms": 411,
+        "arxiv cs ar": 403,
+        "arxiv cs oh": 393,
+        "arxiv cs os": 127,
+        "arxiv cs gl": 29
+      },
+      "isolated_nodes": 0,
+      "feature_columns": 128
+    }
+  },
+  "title": "ogbn-arxiv",
+  "band": "large",
+  "label_type": "semantic",
+  "task": "arXiv subject-area classification",
+  "license": "ODC-BY",
+  "citation": "Hu, Fey, Zitnik, Dong, Ren, Liu, Catasta, Leskovec. Open Graph Benchmark. NeurIPS 2020.",
+  "links": [
+    "https://ogb.stanford.edu/docs/nodeprop/#ogbn-arxiv"
+  ]
+}

polblogs/card.md

@@ -0,0 +1,32 @@
+# Political Blogs (Polblogs)
+
+**Task**: Political-leaning classification  
+**Size band**: small · **Label type**: semantic  
+**Label column**: `leaning` · **Converter**: `converters/convert_small_text.py`
+
+Hyperlink network of US political blogs captured before the 2004 election. Label is the blog's political leaning (0 = liberal, 1 = conservative). Community structure correlates strongly with the label.
+
+## Converted graphs (neext/)
+
+| graph | nodes | edges | classes | feature cols | isolated | class counts |
+|---|---|---|---|---|---|---|
+| default | 1,490 | 16,715 | 2 | 0 | 266 | 0: 758, 1: 732 |
+
+*Conversion notes*: Netzschleuder CSV bundle; 'value' column (0=liberal, 1=conservative) used as label; blog names dropped; directed hyperlinks symmetrized. Contains isolated nodes.
+
+## Source
+
+- [polblogs.csv.zip](https://networks.skewed.de/net/polblogs/files/polblogs.csv.zip) — 93,843 bytes, sha256 `a854433122316724…`, fetched 2026-07-23
+
+**License**: Research use; cite Adamic & Glance  
+**Citation**: Adamic, Glance. The Political Blogosphere and the 2004 U.S. Election: Divided They Blog. LinkKDD 2005.
+
+- https://networks.skewed.de/net/polblogs
+
+## Caveats
+
+- ~266 isolated nodes; NEExT's filter_largest_component or explicit nodes.csv handling decides their fate.
+- Newman's UMich mirror blocks scripted access; Netzschleuder is the reliable source.
+
+---
+*Generated by converters/make_cards.py; stats from metadata.json. Raw files: `source/`. NEExT tables: `neext/`.*

polblogs/metadata.json

@@ -0,0 +1,42 @@
+{
+  "name": "polblogs",
+  "sources": [
+    {
+      "url": "https://networks.skewed.de/net/polblogs/files/polblogs.csv.zip",
+      "final_url": "https://networks.skewed.de/net/polblogs/files/polblogs.csv.zip",
+      "filename": "polblogs.csv.zip",
+      "bytes": 93843,
+      "sha256": "a8544331223167241b7d250d44ceadb140a46b3bee85dd9eb2f55a2ff83bc2d1",
+      "downloaded_at": "2026-07-23T02:14:14+00:00"
+    }
+  ],
+  "conversion": {
+    "script": "convert_small_text.py",
+    "label_column": "leaning",
+    "notes": "Netzschleuder CSV bundle; 'value' column (0=liberal, 1=conservative) used as label; blog names dropped; directed hyperlinks symmetrized. Contains isolated nodes.",
+    "parquet": true
+  },
+  "graphs": {
+    "default": {
+      "nodes": 1490,
+      "edges": 16715,
+      "label_column": "leaning",
+      "classes": 2,
+      "class_counts": {
+        "0": 758,
+        "1": 732
+      },
+      "isolated_nodes": 266,
+      "feature_columns": 0
+    }
+  },
+  "title": "Political Blogs (Polblogs)",
+  "band": "small",
+  "label_type": "semantic",
+  "task": "Political-leaning classification",
+  "license": "Research use; cite Adamic & Glance",
+  "citation": "Adamic, Glance. The Political Blogosphere and the 2004 U.S. Election: Divided They Blog. LinkKDD 2005.",
+  "links": [
+    "https://networks.skewed.de/net/polblogs"
+  ]
+}

pubmed/card.md

@@ -0,0 +1,31 @@
+# Pubmed (Diabetes)
+
+**Task**: Paper topic classification (citation network)  
+**Size band**: medium · **Label type**: semantic  
+**Label column**: `diabetes_type` · **Converter**: `converters/convert_linqs.py`
+
+Citation network of PubMed diabetes papers with 3 classes (diabetes type) and 500 TF-IDF word features densified from the sparse LINQS format. The largest of the classic Planetoid trio.
+
+## Converted graphs (neext/)
+
+| graph | nodes | edges | classes | feature cols | isolated | class counts |
+|---|---|---|---|---|---|---|
+| default | 19,717 | 44,324 | 3 | 500 | 0 | 2: 7,875, 3: 7,739, 1: 4,103 |
+
+*Conversion notes*: Sparse TF-IDF fields densified to 500 w_* columns; labels 1-3 (diabetes type); PubMed IDs remapped (id_mapping.csv).
+
+## Source
+
+- [Pubmed-Diabetes.tgz](https://linqs-data.soe.ucsc.edu/public/Pubmed-Diabetes.tgz) — 14,710,584 bytes, sha256 `218e3c86146cf6d4…`, fetched 2026-07-23
+
+**License**: LINQS research distribution  
+**Citation**: Namata, London, Getoor, Huang. Query-Driven Active Surveying for Collective Classification. MLG 2012.
+
+- https://linqs.org/datasets/
+
+## Caveats
+
+- Old LINQS /lbc/Pubmed* path is dead; /public/Pubmed-Diabetes.tgz is the working source.
+
+---
+*Generated by converters/make_cards.py; stats from metadata.json. Raw files: `source/`. NEExT tables: `neext/`.*

pubmed/metadata.json

@@ -0,0 +1,43 @@
+{
+  "name": "pubmed",
+  "sources": [
+    {
+      "url": "https://linqs-data.soe.ucsc.edu/public/Pubmed-Diabetes.tgz",
+      "final_url": "https://linqs-data.soe.ucsc.edu/public/Pubmed-Diabetes.tgz",
+      "filename": "Pubmed-Diabetes.tgz",
+      "bytes": 14710584,
+      "sha256": "218e3c86146cf6d470b305d824f26c86845f0c858ad3fdd269f6cd2ce63e12bd",
+      "downloaded_at": "2026-07-23T02:14:20+00:00"
+    }
+  ],
+  "conversion": {
+    "script": "convert_linqs.py",
+    "label_column": "diabetes_type",
+    "notes": "Sparse TF-IDF fields densified to 500 w_* columns; labels 1-3 (diabetes type); PubMed IDs remapped (id_mapping.csv).",
+    "parquet": true
+  },
+  "graphs": {
+    "default": {
+      "nodes": 19717,
+      "edges": 44324,
+      "label_column": "diabetes_type",
+      "classes": 3,
+      "class_counts": {
+        "2": 7875,
+        "3": 7739,
+        "1": 4103
+      },
+      "isolated_nodes": 0,
+      "feature_columns": 500
+    }
+  },
+  "title": "Pubmed (Diabetes)",
+  "band": "medium",
+  "label_type": "semantic",
+  "task": "Paper topic classification (citation network)",
+  "license": "LINQS research distribution",
+  "citation": "Namata, London, Getoor, Huang. Query-Driven Active Surveying for Collective Classification. MLG 2012.",
+  "links": [
+    "https://linqs.org/datasets/"
+  ]
+}

reddit-graphsage/card.md

@@ -0,0 +1,32 @@
+# Reddit (GraphSAGE)
+
+**Task**: Subreddit classification of posts  
+**Size band**: large · **Label type**: semantic  
+**Label column**: `subreddit` · **Converter**: `converters/convert_large.py`
+
+Post-post graph where posts are linked when the same user commented on both; the label is the post's subreddit (41 classes). The classic large-scale inductive benchmark and this catalog's scalability stress test.
+
+## Converted graphs (neext/)
+
+| graph | nodes | edges | classes | feature cols | isolated | class counts |
+|---|---|---|---|---|---|---|
+| default | 232,965 | 11,606,919 | 41 | 0 | 1,522 | 15: 28,272, 3: 15,181, 18: 13,999, 0: 13,101, 38: 12,797, 23: 12,146, … |
+
+*Conversion notes*: GraphSAGE node-link json streamed as bytes (links regex-extracted); 41 subreddit classes as ints; label-only nodes.csv (602-dim GloVe features left in source zip).
+
+## Source
+
+- [reddit.zip](https://snap.stanford.edu/graphsage/reddit.zip) — 1,308,432,264 bytes, sha256 `25337a21540cd373…`, fetched 2026-07-23
+
+**License**: SNAP research use; cite GraphSAGE  
+**Citation**: Hamilton, Ying, Leskovec. Inductive Representation Learning on Large Graphs. NeurIPS 2017.
+
+- https://snap.stanford.edu/graphsage/
+
+## Caveats
+
+- 233k nodes / 11.6M edges and dense — egonet decomposition only with small sample_fraction.
+- Label-only nodes.csv by design; 602-dim GloVe features remain in source/reddit.zip.
+
+---
+*Generated by converters/make_cards.py; stats from metadata.json. Raw files: `source/`. NEExT tables: `neext/`.*

reddit-graphsage/metadata.json

@@ -0,0 +1,83 @@
+{
+  "name": "reddit-graphsage",
+  "sources": [
+    {
+      "url": "https://snap.stanford.edu/graphsage/reddit.zip",
+      "final_url": "https://snap.stanford.edu/graphsage/reddit.zip",
+      "filename": "reddit.zip",
+      "bytes": 1308432264,
+      "sha256": "25337a21540cd373e4cee3751e6600324ab6a7377ef3966bb49f57412a17ed02",
+      "downloaded_at": "2026-07-23T02:16:38+00:00"
+    }
+  ],
+  "conversion": {
+    "script": "convert_large.py",
+    "label_column": "subreddit",
+    "nodes_in_source_json": 231443,
+    "notes": "GraphSAGE node-link json streamed as bytes (links regex-extracted); 41 subreddit classes as ints; label-only nodes.csv (602-dim GloVe features left in source zip).",
+    "parquet": true,
+    "parquet_notes": "nodes.parquet includes the full feature matrix omitted from nodes.csv for size."
+  },
+  "graphs": {
+    "default": {
+      "nodes": 232965,
+      "edges": 11606919,
+      "label_column": "subreddit",
+      "classes": 41,
+      "class_counts": {
+        "15": 28272,
+        "3": 15181,
+        "18": 13999,
+        "0": 13101,
+        "38": 12797,
+        "23": 12146,
+        "8": 11187,
+        "19": 10308,
+        "22": 8222,
+        "27": 5962,
+        "40": 5112,
+        "29": 5101,
+        "33": 4960,
+        "10": 4928,
+        "14": 4854,
+        "28": 4673,
+        "31": 4570,
+        "26": 4239,
+        "37": 4233,
+        "35": 4202,
+        "36": 4180,
+        "21": 4066,
+        "6": 3952,
+        "5": 3597,
+        "1": 3550,
+        "34": 3429,
+        "2": 3302,
+        "39": 3099,
+        "11": 2964,
+        "30": 2846,
+        "13": 2731,
+        "17": 2639,
+        "4": 2322,
+        "9": 2246,
+        "7": 2138,
+        "12": 1696,
+        "25": 1659,
+        "20": 1596,
+        "32": 1575,
+        "16": 1003,
+        "24": 328
+      },
+      "isolated_nodes": 1522,
+      "feature_columns": 0
+    }
+  },
+  "title": "Reddit (GraphSAGE)",
+  "band": "large",
+  "label_type": "semantic",
+  "task": "Subreddit classification of posts",
+  "license": "SNAP research use; cite GraphSAGE",
+  "citation": "Hamilton, Ying, Leskovec. Inductive Representation Learning on Large Graphs. NeurIPS 2017.",
+  "links": [
+    "https://snap.stanford.edu/graphsage/"
+  ]
+}

roman-empire/card.md

@@ -0,0 +1,27 @@
+# Roman Empire
+
+**Task**: Syntactic-role classification  
+**Size band**: medium · **Label type**: structural  
+**Label column**: `syntactic_role` · **Converter**: `converters/convert_npz.py`
+
+Graph built from the 'Roman Empire' Wikipedia article: nodes are words in text order, edges connect consecutive or syntactically dependent words. Label is the word's syntactic role (18 classes). Chain-like, extremely sparse — the anti-homophily extreme where local structure carries the signal.
+
+## Converted graphs (neext/)
+
+| graph | nodes | edges | classes | feature cols | isolated | class counts |
+|---|---|---|---|---|---|---|
+| default | 22,662 | 32,927 | 18 | 300 | 0 | 1: 3,163, 2: 3,133, 3: 2,502, 4: 2,487, 17: 2,194, 5: 1,359, … |
+
+*Conversion notes*: yandex heterophilous-graphs npz; edges symmetrized; 300 f_* feature columns.
+
+## Source
+
+- [roman_empire.npz](https://raw.githubusercontent.com/yandex-research/heterophilous-graphs/main/data/roman_empire.npz) — 20,401,489 bytes, sha256 `a58ba741d123bf89…`, fetched 2026-07-23
+
+**License**: MIT (yandex-research)  
+**Citation**: Platonov, Kuznedelev, Diskin, Babenko, Prokhorenkova. A Critical Look at the Evaluation of GNNs under Heterophily. ICLR 2023.
+
+- https://github.com/yandex-research/heterophilous-graphs
+
+---
+*Generated by converters/make_cards.py; stats from metadata.json. Raw files: `source/`. NEExT tables: `neext/`.*

roman-empire/metadata.json

@@ -0,0 +1,58 @@
+{
+  "name": "roman-empire",
+  "sources": [
+    {
+      "url": "https://raw.githubusercontent.com/yandex-research/heterophilous-graphs/main/data/roman_empire.npz",
+      "final_url": "https://raw.githubusercontent.com/yandex-research/heterophilous-graphs/main/data/roman_empire.npz",
+      "filename": "roman_empire.npz",
+      "bytes": 20401489,
+      "sha256": "a58ba741d123bf892fe5c872138d07463d75a2e9012360b8dd78ac2d4766d428",
+      "downloaded_at": "2026-07-23T02:14:51+00:00"
+    }
+  ],
+  "conversion": {
+    "script": "convert_npz.py",
+    "label_column": "syntactic_role",
+    "notes": "yandex heterophilous-graphs npz; edges symmetrized; 300 f_* feature columns.",
+    "parquet": true
+  },
+  "graphs": {
+    "default": {
+      "nodes": 22662,
+      "edges": 32927,
+      "label_column": "syntactic_role",
+      "classes": 18,
+      "class_counts": {
+        "1": 3163,
+        "2": 3133,
+        "3": 2502,
+        "4": 2487,
+        "17": 2194,
+        "5": 1359,
+        "6": 1244,
+        "7": 1080,
+        "0": 944,
+        "8": 852,
+        "9": 789,
+        "10": 717,
+        "11": 445,
+        "12": 428,
+        "13": 365,
+        "14": 329,
+        "15": 319,
+        "16": 312
+      },
+      "isolated_nodes": 0,
+      "feature_columns": 300
+    }
+  },
+  "title": "Roman Empire",
+  "band": "medium",
+  "label_type": "structural",
+  "task": "Syntactic-role classification",
+  "license": "MIT (yandex-research)",
+  "citation": "Platonov, Kuznedelev, Diskin, Babenko, Prokhorenkova. A Critical Look at the Evaluation of GNNs under Heterophily. ICLR 2023.",
+  "links": [
+    "https://github.com/yandex-research/heterophilous-graphs"
+  ]
+}

tolokers/card.md

@@ -0,0 +1,31 @@
+# Tolokers
+
+**Task**: Banned crowdworker prediction  
+**Size band**: medium · **Label type**: anomaly  
+**Label column**: `is_banned` · **Converter**: `converters/convert_npz.py`
+
+Workers of the Toloka crowdsourcing platform connected when they worked on the same tasks; the label marks workers who were banned (21.8% positive — a workable imbalance). Dense graph (avg degree ~88).
+
+## Converted graphs (neext/)
+
+| graph | nodes | edges | classes | feature cols | isolated | class counts |
+|---|---|---|---|---|---|---|
+| default | 11,758 | 519,000 | 2 | 10 | 0 | 0: 9,192, 1: 2,566 |
+
+*Conversion notes*: yandex heterophilous-graphs npz; edges symmetrized; 10 f_* feature columns.
+
+## Source
+
+- [tolokers.npz](https://raw.githubusercontent.com/yandex-research/heterophilous-graphs/main/data/tolokers.npz) — 1,329,769 bytes, sha256 `dacf3ac94cec53d0…`, fetched 2026-07-23
+
+**License**: MIT (yandex-research)  
+**Citation**: Platonov et al. ICLR 2023 (data from Toloka).
+
+- https://github.com/yandex-research/heterophilous-graphs
+
+## Caveats
+
+- Density makes k_hop=2 egonets large; start at k_hop=1.
+
+---
+*Generated by converters/make_cards.py; stats from metadata.json. Raw files: `source/`. NEExT tables: `neext/`.*

tolokers/metadata.json

@@ -0,0 +1,42 @@
+{
+  "name": "tolokers",
+  "sources": [
+    {
+      "url": "https://raw.githubusercontent.com/yandex-research/heterophilous-graphs/main/data/tolokers.npz",
+      "final_url": "https://raw.githubusercontent.com/yandex-research/heterophilous-graphs/main/data/tolokers.npz",
+      "filename": "tolokers.npz",
+      "bytes": 1329769,
+      "sha256": "dacf3ac94cec53d03cd2adb5255c08b33dee1656c33ca8164a464bd9450a1667",
+      "downloaded_at": "2026-07-23T02:14:53+00:00"
+    }
+  ],
+  "conversion": {
+    "script": "convert_npz.py",
+    "label_column": "is_banned",
+    "notes": "yandex heterophilous-graphs npz; edges symmetrized; 10 f_* feature columns.",
+    "parquet": true
+  },
+  "graphs": {
+    "default": {
+      "nodes": 11758,
+      "edges": 519000,
+      "label_column": "is_banned",
+      "classes": 2,
+      "class_counts": {
+        "0": 9192,
+        "1": 2566
+      },
+      "isolated_nodes": 0,
+      "feature_columns": 10
+    }
+  },
+  "title": "Tolokers",
+  "band": "medium",
+  "label_type": "anomaly",
+  "task": "Banned crowdworker prediction",
+  "license": "MIT (yandex-research)",
+  "citation": "Platonov et al. ICLR 2023 (data from Toloka).",
+  "links": [
+    "https://github.com/yandex-research/heterophilous-graphs"
+  ]
+}

weibo/card.md

@@ -0,0 +1,32 @@
+# Weibo (PyGOD)
+
+**Task**: Social spam detection  
+**Size band**: medium · **Label type**: anomaly  
+**Label column**: `is_outlier` · **Converter**: `converters/convert_pygod.py`
+
+Sina Weibo user-user graph (shared-hashtag edges) with 400 post-derived features and organic spammer labels (10.3% anomalous). Real social misbehavior at Workbench-comfortable scale, but dense (avg degree ~90).
+
+## Converted graphs (neext/)
+
+| graph | nodes | edges | classes | feature cols | isolated | class counts |
+|---|---|---|---|---|---|---|
+| default | 8,405 | 377,271 | 2 | 400 | 0 | 0: 8,058, 1: 347 |
+
+*Conversion notes*: PyGOD .pt (torch pickle, trusted pygod-team/data source) via stub unpickling; edge_index symmetrized; 400 x_* feature columns.
+
+## Source
+
+- [weibo.pt.zip](https://raw.githubusercontent.com/pygod-team/data/main/weibo.pt.zip) — 13,339,024 bytes, sha256 `c4a5fe4ca61a9566…`, fetched 2026-07-23
+
+**License**: MIT (pygod-team/data)  
+**Citation**: Liu et al. BOND. NeurIPS 2022 D&B (data: Zhao et al.).
+
+- https://github.com/pygod-team/data
+
+## Caveats
+
+- Dense — k_hop=1 with sampling first.
+- Torch-pickle source; stub-unpickled from the hash-verified official repo.
+
+---
+*Generated by converters/make_cards.py; stats from metadata.json. Raw files: `source/`. NEExT tables: `neext/`.*

weibo/metadata.json

@@ -0,0 +1,42 @@
+{
+  "name": "weibo",
+  "sources": [
+    {
+      "url": "https://raw.githubusercontent.com/pygod-team/data/main/weibo.pt.zip",
+      "final_url": "https://raw.githubusercontent.com/pygod-team/data/main/weibo.pt.zip",
+      "filename": "weibo.pt.zip",
+      "bytes": 13339024,
+      "sha256": "c4a5fe4ca61a9566e051bf0544816c9340c49b031dfd07693b473d18e4e6f2f9",
+      "downloaded_at": "2026-07-23T02:14:55+00:00"
+    }
+  ],
+  "conversion": {
+    "script": "convert_pygod.py",
+    "label_column": "is_outlier",
+    "notes": "PyGOD .pt (torch pickle, trusted pygod-team/data source) via stub unpickling; edge_index symmetrized; 400 x_* feature columns.",
+    "parquet": true
+  },
+  "graphs": {
+    "default": {
+      "nodes": 8405,
+      "edges": 377271,
+      "label_column": "is_outlier",
+      "classes": 2,
+      "class_counts": {
+        "0": 8058,
+        "1": 347
+      },
+      "isolated_nodes": 0,
+      "feature_columns": 400
+    }
+  },
+  "title": "Weibo (PyGOD)",
+  "band": "medium",
+  "label_type": "anomaly",
+  "task": "Social spam detection",
+  "license": "MIT (pygod-team/data)",
+  "citation": "Liu et al. BOND. NeurIPS 2022 D&B (data: Zhao et al.).",
+  "links": [
+    "https://github.com/pygod-team/data"
+  ]
+}

wikics/card.md

@@ -0,0 +1,27 @@
+# WikiCS
+
+**Task**: CS article branch classification  
+**Size band**: medium · **Label type**: semantic  
+**Label column**: `category` · **Converter**: `converters/convert_wikics.py`
+
+Wikipedia computer-science article network with 10 branch classes and 300-dim mean-GloVe features. Harder and more modern than the Planetoid trio, with canonical splits available in the source JSON.
+
+## Converted graphs (neext/)
+
+| graph | nodes | edges | classes | feature cols | isolated | class counts |
+|---|---|---|---|---|---|---|
+| default | 11,701 | 215,603 | 10 | 300 | 337 | 4: 2,679, 2: 2,153, 3: 1,933, 9: 1,424, 7: 865, 5: 780, … |
+
+*Conversion notes*: Adjacency lists flattened and symmetrized; 10 CS-branch classes; 300 mean-GloVe glove_* feature columns.
+
+## Source
+
+- [data.json](https://raw.githubusercontent.com/pmernyei/wiki-cs-dataset/master/dataset/data.json) — 82,533,647 bytes, sha256 `9bf8cb3ef8eeae81…`, fetched 2026-07-23
+
+**License**: MIT (dataset code); article text CC BY-SA  
+**Citation**: Mernyei, Cangea. Wiki-CS: A Wikipedia-Based Benchmark for Graph Neural Networks. GRL+ 2020.
+
+- https://github.com/pmernyei/wiki-cs-dataset
+
+---
+*Generated by converters/make_cards.py; stats from metadata.json. Raw files: `source/`. NEExT tables: `neext/`.*

wikics/metadata.json

@@ -0,0 +1,50 @@
+{
+  "name": "wikics",
+  "sources": [
+    {
+      "url": "https://raw.githubusercontent.com/pmernyei/wiki-cs-dataset/master/dataset/data.json",
+      "final_url": "https://raw.githubusercontent.com/pmernyei/wiki-cs-dataset/master/dataset/data.json",
+      "filename": "data.json",
+      "bytes": 82533647,
+      "sha256": "9bf8cb3ef8eeae81b25e6ccbe0ea195600c205d7edf63ce04f2ec8d9c7dcb3d8",
+      "downloaded_at": "2026-07-23T02:14:46+00:00"
+    }
+  ],
+  "conversion": {
+    "script": "convert_wikics.py",
+    "label_column": "category",
+    "notes": "Adjacency lists flattened and symmetrized; 10 CS-branch classes; 300 mean-GloVe glove_* feature columns.",
+    "parquet": true
+  },
+  "graphs": {
+    "default": {
+      "nodes": 11701,
+      "edges": 215603,
+      "label_column": "category",
+      "classes": 10,
+      "class_counts": {
+        "4": 2679,
+        "2": 2153,
+        "3": 1933,
+        "9": 1424,
+        "7": 865,
+        "5": 780,
+        "1": 667,
+        "8": 492,
+        "6": 413,
+        "0": 295
+      },
+      "isolated_nodes": 337,
+      "feature_columns": 300
+    }
+  },
+  "title": "WikiCS",
+  "band": "medium",
+  "label_type": "semantic",
+  "task": "CS article branch classification",
+  "license": "MIT (dataset code); article text CC BY-SA",
+  "citation": "Mernyei, Cangea. Wiki-CS: A Wikipedia-Based Benchmark for Graph Neural Networks. GRL+ 2020.",
+  "links": [
+    "https://github.com/pmernyei/wiki-cs-dataset"
+  ]
+}

wikipedia-articles/card.md

@@ -0,0 +1,34 @@
+# Wikipedia Article Networks (Chameleon / Squirrel / Crocodile)
+
+**Task**: Traffic-level classification (binned)  
+**Size band**: small-medium · **Label type**: semantic  
+**Label column**: `traffic_quintile` · **Converter**: `converters/convert_musae.py`
+
+Wikipedia page-page networks on three topics. The raw target is continuous monthly traffic; following the geom-gcn convention it is binned into equal-frequency quintiles (5 classes). Chameleon/Squirrel are standard heterophily benchmarks in binned form.
+
+## Converted graphs (neext/)
+
+| graph | nodes | edges | classes | feature cols | isolated | class counts |
+|---|---|---|---|---|---|---|
+| chameleon | 2,277 | 31,371 | 5 | 0 | 0 | 3: 521, 1: 460, 0: 456, 2: 453, 4: 387 |
+| crocodile | 11,631 | 170,773 | 5 | 0 | 0 | 0: 2,465, 4: 2,326, 3: 2,325, 2: 2,315, 1: 2,200 |
+| squirrel | 5,201 | 198,353 | 5 | 0 | 0 | 0: 1,042, 1: 1,040, 3: 1,040, 4: 1,040, 2: 1,039 |
+
+*Conversion notes*: Raw target is continuous monthly traffic; binned to equal-frequency quintiles (0-4) following the geom-gcn 5-class convention. Feature JSONs not columnized.
+
+## Source
+
+- [wikipedia.zip](https://snap.stanford.edu/data/wikipedia.zip) — 3,784,155 bytes, sha256 `aeef409fca6b08ab…`, fetched 2026-07-23
+
+**License**: SNAP / MUSAE; cite MUSAE  
+**Citation**: Rozemberczki, Allen, Sarkar. MUSAE. 2021.
+
+- https://snap.stanford.edu/data/wikipedia-article-networks.html
+
+## Caveats
+
+- Quintile binning is our reproduction of the geom-gcn convention (equal-frequency qcut on raw traffic), not an official label file.
+- Known duplicate-node criticism of Squirrel/Chameleon in the literature (Platonov et al. 2023).
+
+---
+*Generated by converters/make_cards.py; stats from metadata.json. Raw files: `source/`. NEExT tables: `neext/`.*

wikipedia-articles/metadata.json

@@ -0,0 +1,75 @@
+{
+  "name": "wikipedia-articles",
+  "sources": [
+    {
+      "url": "https://snap.stanford.edu/data/wikipedia.zip",
+      "final_url": "https://snap.stanford.edu/data/wikipedia.zip",
+      "filename": "wikipedia.zip",
+      "bytes": 3784155,
+      "sha256": "aeef409fca6b08abe9e3751bf41839b2544d499f0eca174701f87c9f93c71b86",
+      "downloaded_at": "2026-07-23T02:14:33+00:00"
+    }
+  ],
+  "conversion": {
+    "script": "convert_musae.py",
+    "label_column": "traffic_quintile",
+    "notes": "Raw target is continuous monthly traffic; binned to equal-frequency quintiles (0-4) following the geom-gcn 5-class convention. Feature JSONs not columnized.",
+    "parquet": true
+  },
+  "graphs": {
+    "chameleon": {
+      "nodes": 2277,
+      "edges": 31371,
+      "label_column": "traffic_quintile",
+      "classes": 5,
+      "class_counts": {
+        "3": 521,
+        "1": 460,
+        "0": 456,
+        "2": 453,
+        "4": 387
+      },
+      "isolated_nodes": 0,
+      "feature_columns": 0
+    },
+    "crocodile": {
+      "nodes": 11631,
+      "edges": 170773,
+      "label_column": "traffic_quintile",
+      "classes": 5,
+      "class_counts": {
+        "0": 2465,
+        "4": 2326,
+        "3": 2325,
+        "2": 2315,
+        "1": 2200
+      },
+      "isolated_nodes": 0,
+      "feature_columns": 0
+    },
+    "squirrel": {
+      "nodes": 5201,
+      "edges": 198353,
+      "label_column": "traffic_quintile",
+      "classes": 5,
+      "class_counts": {
+        "0": 1042,
+        "1": 1040,
+        "3": 1040,
+        "4": 1040,
+        "2": 1039
+      },
+      "isolated_nodes": 0,
+      "feature_columns": 0
+    }
+  },
+  "title": "Wikipedia Article Networks (Chameleon / Squirrel / Crocodile)",
+  "band": "small-medium",
+  "label_type": "semantic",
+  "task": "Traffic-level classification (binned)",
+  "license": "SNAP / MUSAE; cite MUSAE",
+  "citation": "Rozemberczki, Allen, Sarkar. MUSAE. 2021.",
+  "links": [
+    "https://snap.stanford.edu/data/wikipedia-article-networks.html"
+  ]
+}